Characterizing the pore structure of a porous building material: Impacts of image acquisition, image denoising and image segmentation choices

The porous nature of a building material significantly influences its functional properties, with the pore structure being particularly crucial. Imaging techniques are widely used to characterize pore structures, typically involving image acquisition, image denoising, and image segmentation. This paper analyses the distinct impacts of these procedures on the resulting pore structure characterization, by application on ceramic brick, a commonly applied building material. To visualize its extensive pore size spectrum, a coherent image chain comprising eight sets of three-dimensional greyscale images was acquired. To that aim, two imaging methods were used: X-ray computed tomography and X-ray microscopy, with voxel sizes spanning from nanometers to micrometers. The acquired images were firstly denoised – to enhance their quality – by applying various filters. They were subsequently segmented – to distinguish pores from matrix – employing three thresholding-based techniques and two AI-assisted approaches. And ultimately the resulting pore networks were extracted with the Maximal Ball method. The impacts of image acquisition settings, image denoising filters, and image segmentation approaches were first evaluated qualitatively on two-dimensional image slices and then quantitatively through pore-network-based evaluation metrics. Our findings show that capturing the broad and smooth pore size distribution typical of porous building materials requires multiple image acquisitions with varying voxel sizes, to resolve the usual conflict between field of view and resolution. Combining a median filter with a non-local means filter effectively reduces noise and artifacts for this application. However, filter selection should be tailored to each specific scenario, to minimize noise while preserving critical details. While different segmentation methods have a minimal effect on the overall pore structure topology, the use of hysteresis thresholding combined with a top-hat transform yields the most faithful representation of the pore structure. Complementarily, the AI-assisted segmentation has the potential to provide reliable results even with minimal image denoising.